|Abstract:||This paper focuses on the elaboration of inputs, drawing on data from two patterns of cluster reduction in early English, Dutch and German. In the sonority pattern, the least sonorous segment survives; in the head pattern, the structurally-defined head of the cluster survives. Earlier research has shown that the sonority pattern can be arrived at using constraints that assess the sonority value of onsets only. This is not so for the head pattern: in /S/+sonorant clusters, there is a mismatch between cluster head and least sonorous segment. In this paper, a unified analysis is proposed for both patterns. It begins by adopting a structural difference between obstruent- versus /S/-initial clusters: the former are left-headed branching onsets; the latter contain an initial appendix. It is argued that faithfulness constraints must reference heads of constituents; MaxHead(Onset) plays a crucial role in the analysis. MaxHead requires inputs to be fully prosodified. This, in turn, demands knowledge of the structures that are permitted in the target language. Accordingly, it is proposed that children\'s inputs are initially prosodified for simplex onsets and rhymes, that is, for constituent heads only. While there is often a correlation between the head of a cluster and low sonority, it is shown that heads cannot be determined solely on the basis of relative prominence; distributional evidence must be factored in and understanding this requires knowledge that is relatively sophisticated. It is proposed that children initially make decisions about headedness on the basis of sonority until the distributional facts are understood. It is for this reason that two patterns of cluster reduction are observed: the sonority pattern represents stage 1; the head pattern, stage 2.